Analytic Representation vs. Angle Modulation of Hilbert Transform of Fast Walsh-Hadamard Coefficients (HTFWHC) in Epileptic EEG Classification

The demand for automation of epilepsy diagnosis from electroencephalogram (EEG) signals as an alternative to manual interpretation performed by a human specialist has been increasing in recent years. This paper describes a novel class of hybrid features for categorizing EEG signals of normal and epileptic patients. The proposed hybrid feature benefits from both frequency and nonlinear information concealed in the signal. The former implemented the fast Walsh-Hadamard transform, and the latter used the Hilbert transform (HTFWHT). The analytic representation (AR) and angle modulation (AM) of the hybrid feature is formed and characterized by several measures of central tendency and dispersion. The efficiency of AR and AM indices in a multi-class epilepsy diagnosis is evaluated using the probabilistic neural network. Bonn EEG database of healthy and epileptic patients, during seizure occurrence and seizure-free, was assessed. AR outperformed the AM measures of the HTFWTC. The system has detected all classes with the highest rates of 100% and 95% using AR and AM measures of the HTFWTC, respectively. The proposed approach achieved a higher accuracy rate compared to the state-of-art algorithms.